Every figure below is measured out of sample on a rolling-origin backtest at 3 past dates, and shown beside the benchmark anyone can compute: last year's same month on the same routes.
| Months ahead | Model | Last year | Model wins | 90th pct | 95% band holds | Routes |
|---|---|---|---|---|---|---|
| 1 | 12% | 17% | 65% | 53% | 97% | 28,632 |
| 2 | 17% | 17% | 56% | 71% | 96% | 28,567 |
| 3 | 16% | 16% | 60% | 72% | 96% | 28,360 |
| 4 | 17% | 17% | 57% | 87% | 96% | 27,751 |
| 5 | 18% | 18% | 56% | 94% | 95% | 27,308 |
| 6 | 17% | 17% | 57% | 79% | 96% | 27,459 |
| 7 | 18% | 18% | 53% | 88% | 96% | 21,767 |
| 8 | 18% | 17% | 47% | 86% | 96% | 21,710 |
| 9 | 19% | 18% | 49% | 92% | 95% | 21,485 |
| 10 | 29% | 25% | 42% | 119% | 96% | 21,006 |
| 11 | 30% | 25% | 43% | 114% | 95% | 20,730 |
| 12 | 29% | 24% | 45% | 100% | 96% | 21,211 |
Typical route error (passengers), median across routes. Last year: last-year-same-month on the same routes. 95% band holds: share of actuals that fell inside the published interval.
| Months ahead | Model | Last year | Model wins | 90th pct | 95% band holds | Routes |
|---|---|---|---|---|---|---|
| 1 | 32% | 46% | 67% | 303% | 95% | 15,709 |
| 2 | 41% | 46% | 60% | 350% | 95% | 15,682 |
| 3 | 41% | 47% | 60% | 321% | 95% | 15,622 |
| 4 | 46% | 47% | 59% | 369% | 95% | 15,482 |
| 5 | 47% | 48% | 58% | 379% | 95% | 15,507 |
| 6 | 47% | 45% | 55% | 306% | 95% | 15,658 |
| 7 | 53% | 50% | 54% | 336% | 95% | 13,771 |
| 8 | 52% | 49% | 53% | 350% | 95% | 13,724 |
| 9 | 53% | 50% | 53% | 339% | 94% | 13,706 |
| 10 | 57% | 62% | 60% | 533% | 95% | 13,565 |
| 11 | 55% | 61% | 60% | 811% | 95% | 13,501 |
| 12 | 56% | 60% | 59% | 728% | 94% | 13,636 |
Typical route error (tonnes), median across routes. Last year: last-year-same-month on the same routes. 95% band holds: share of actuals that fell inside the published interval.
Cargo tonnage is far more volatile than passengers; the 90th percentile shows it. The gap to the benchmark is larger than for passengers at every horizon.
Scored against what airlines actually operated, 2024-07 to 2026-06, across 639,045 passenger and 87,113 cargo route-months.
Load factor achieved at the recommended frequency, passenger routes, by the label the platform gave the route:
If the label carried no information, the three figures would be the same.
These are the all-routes figures. On the platform the same measurements are broken down by region, haul length, domestic/international, route size, history available and data source, plus error spread, direction of bias, stability between updates and peak-month accuracy — and every route carries its own measured error against comparable routes.
See whether it works for your market →